Thij, T. ten (2004) Merging shared action categories in physical robotic agents through imitation. Master's Thesis / Essay, Artificial Intelligence.
|
Text
AI_Ma_2004_TtenTij.CV.pdf - Published Version Download (4MB) | Preview |
Abstract
Artificial intelligence is concerned with studying and recreating human intelligence. Humans distinguish themselves from other animals and primates through their use of a rich and expressive communication system: natural language. The enterprise of studying human intelligence can not be complete without studying human natural language. When one is confronted with the efforts of linguists over the last decades to capture our knowledge of language, it becomes clear that this is not as easy as it may have seemed. Linguists have been searching for a set of rules that determine whether a sentence is grammatical or not. As yet no such set has been found that is capable of describing even one human natural language. Some linguists claim that the fact that such a ruleset has not been found by intelligent researches proves that children are born with a 'language module' (aka universal grammar). Luc Steels' Artificial Intelligence Lab in Brussel has been trying to use computer models to investigate what aspects of language can be explained without assuming that they have to be innate. They try to show that language is a self-organizing complex dynamic system that emerges from human cultural interactions. The human brain must possess some mechanisms that other animals do not, but these does not have to be specific for language. Using populations of interacting agents, it has already been shown that some aspects that were assumed to be innate can be explained by cultural transmission (eg vowel systems (Bart de Boer) and colorcategories (Tony Belpaeme)). My research focussed on the cultural transmission of action categories. The experiments have been performed on a robotic arm that uses four degrees of freedom and a stereoscopic camera. Using the arm and camera as embodiment, agents engage in games where they try to imitate the action they saw the other agent perform. The only feedback between agents during the experiment is given by the initiator of the game as one bit of information on whether he judges the game's outcome as successful. In my experiment actions are not coupled to meaning, so it is not a model of the cultural learning of real gestures. But it turns out that a population using a scheme as simple as described is able to develop a shared repertoire of action categories in a very robust manner.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 15 Feb 2018 07:30 |
| Last Modified: | 15 Feb 2018 07:30 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/8980 |
Actions (login required)
 |
View Item |